Hacker News new | past | comments | ask | show | jobs | submit login
Destruction of large numbers of jobs by robots unlikely, says new OECD Study (robotenomics.com)
75 points by Robotenomics on May 19, 2016 | hide | past | web | favorite | 114 comments

Actual OECD study: [1]

The study says that, for the US, 9% of people are at "high risk" of automated out of a job. (That probably means "replaceable right now".) But 38% of people are potentially replaceable.

There's an assumption in the OECD report that the entire job of a human must be replaced. But that's not how automation works. We're seeing this in the more advanced law firms. What used to take a senior attorney, a few junior attorneys, a large number of paralegals, and a big clerical staff can now be done by one attorney, one paralegal, and a lot of software and databases. The OECD study claims that the risk to people with high levels of education is almost nil. Ask any newly graduated lawyer trying to get a job.

That's the fundamental flaw in this study - it assumes one for one replacement. What really happens is that the workflow is restructured so that fewer people and more hardware are involved.

[1] http://www.oecd-ilibrary.org/docserver/download/5jlz9h56dvq7...

One for one replacement is a truly strange standard to hold to. After all, even trucking automation will not be a 100% loss - if nothing else, we'll see a rise in full-service highway gas stations to fuel driverless trucks.

On the other hand, even 10% replacement of an industry (9 workers doing what 10 workers do now) is a big deal, and there are a lot of careers (legal, HR, document handling) that should expect to see values like 75% replacement. I can't think why someone would (in good faith) treat a one-to-one statistic as the whole of the answer here.

If driverless trucking takes off, why would the gas station need to be full service (staffed with humans)?

Compared to building an autonomous truck, autonomous refueling seems like the easy part.

Think of it this way:

Right now, to move one trucks' worth of cargo for one hour requires one man hour. i.e.: 1 man-hour per cargo-hour.

If you have a single human that full-services 200 self-driving trucks in one day, you have: 1 man-hour per 200 cargo-hours.

Sure, autonomous refueling is easier than self-driving, but self-driving trucks also makes the human relatively way cheaper to employ.

My thoughts exactly.

The usual standard for automation, I believe, is 18-24 months salary - if the capital cost can be recouped in two years automation is worthwhile. For hundreds of well-paid truckers that's a relatively generous target, but automating a reliable gas pumping system just to save on one minimum-wage gas pumper looks like a much lower ROI.

I recently spent $100,000 on some automation for my factory. The first $60,000 was for a machine that will take an outside supplier cost of $5000 per month, and replace it with a machine payment of $1000 per month. I now need to supply an operator, space, electrical and consumables, but I save $4000 to do that with. My 'new' worker will be just one of the existing workers (or myself) who will push the buttons on this mostly automated machine. The human still needs to load and unload the machine.

The other $40,000 was spent on NEW machines that automatically do NEW processes which we could never do before. I have a lot of pressure from the market to deliver new goods, and I have a lot of competitors both in the USA and in China and India copying my existing products. So this investment is about new streams of revenue (and I like toys/equipment!) . This is a $500 per month investment on the lease.

I have a reasonable expectation that the new machines could help us achieve a sales doubling in the next three years.

So in my case, 18-24 months salary replacement was never the thought process. It was honestly more about getting rid of a supplier, bringing it in-house to control quality and output, and saving money. I have a lot of automation, and I think the bigger concept most don't get is that the automation is way way more accurate and repeatable than a human. Quality goes up as costs go down. The quality part is a big big part of it.

The factor is availability- you need three of these workers for a around the clock shift. AI does not sleep.

Ok, so you have 3 employees working 24/7 for an average of a $20k salary (33% more than minimum wage). That's $60k/year or $120k over 24 months (based on Bartweiss' ROI estimation).

Do you think it's possible to deliver an automated gas pump that services 4,800 trucks per day (~1 truck every 3 minutes) for $120k? I don't.

I don't think 3 humans can service 4800 trucks per day either.

At least, not without more automation than existing pumps have.

Eventually I think it could happen. It simply requires economies of scale--the hardware is not the expensive part of an automated gas pump.

Yes, its the software. And the software needs data to have the machine learn, given this data, a machine designed for fitting niches, can squeeze itself into even unlikely corners.

Except the employees are going to need breaks in between working and I'm going to just guess that manning a gas station in the middle of desert alone makes you insane pretty fast. So at least double that salary estimate if you truly want to keep things rolling with 1 truck every 3 minutes.

I'm not so sure - presumably these are going to be highway truck stops/gas stations. You're not going to leave one guy there alone and shut everything down when he goes on break, you're going to add another 1-2 people to the daily shift at an existing gas station, and share the extra load across them.

My numbers do break down in that one guy per shift probably can't fuel everything, so we might need to double it. (In states with locking pumps, he probably can, though.)

But again what if he needs to take a shit? I don't know about you, but I can't do my business and wipe well in less than 3 minutes, so the trucks stand there. Then there's the ~30min lunch break, that's 10 trucks just idling in the worst case scenario. Then just plain human interaction.

It's not just a factor if he can possibly fill every truck by himself, there are other human factors to consider. You need at least two men manning the station all the time for the autonomous trucks to refill there + maybe a spare guy for regular folk who come to visit.

Not the first year, but the next pumps come with automated fueling as standard equipment, and we can write off the upgrades over the next few years for tax savings.

Track per 3 minutes is 480 tracks per day.

Rentier economy dynamics now encroaching on human capital.

But you might have more trucks, not necessarily less humans.

Until the automate the process of inspection and refueling it would be a good job to place people. Treat it like an airplane coming in from a flight. Support crew go out to verify the condition the truck thinks its in, fix minor things the truck called out, and refuel it. Sign off on the paperwork stating truck meets all acceptable Federal and State laws.

This is definitely part of it - humans are great at flexible-requirement tasks. Gas the truck, kick the tires, squeegee off the sensors, make sure nothing is obviously falling apart. Rather than paying one human per truck, you pay one human per ~200 trucks to do all of the minor, unpredictable service tasks needed.

Far fewer units to move and a much longer expected life on the hardware. Hiring a fuel filler is an easy retrofit.

It would also likely be prudent to have a human visually inspecting the trucks and able to perform small maintenance tasks. My car can tell me the tire pressure is low, but it can't change the tire. It can tell me the backup sensor is obscured, but it can't clean it. Maybe it didn't even notice it hit something/something hit it (like a fast, low flying bird) and now the fender is hanging off. Things that might need to be done in between distribution centers that would be easier for a human to check, at least in the short term.

Yeah, my guess is we'll see convoys of automated trucks first, with one operator for each line of five or ten trucks. In part for small upkeep and correction in real time, in part to increase the risks of straight up highway robbery... If you can blow out an automated truck's tires and know it will take a day for a repair vehicle to show up, you might as well go for it...

Trucks would be standardized and have sensors to detect many types of damage and the truck could be rerouted to an automated repair facility (or perhaps a repair vehicle would be dispatched). Visual inspection could be done better automatically.

Automation that relies on humans to minimally function is garbage.

There will always be humans overseeing automation for the foreseeable future. At least until we figure out what all will/can go wrong. There are humans supervising automation in factories and warehouses, just in case. It's simply a good idea until we have AI as smart and observant as a human.

By this logic should we have autonomous refueling already?

After factoring in the cost of the robotics I suspect paying humans minimum wage is quite a bit cheaper. Especially given the diverse outdoor environments such a system would need to be maintained in

By this logic should we have autonomous refueling already?

It's been done a few times.[1][2] The systems are rather slow and clunky, but work. Dealing with all the variation between cars runs up the cost.

Tesla has a charging robot.[3] This is much easier, since Tesla controls both sides of the interface. The car actively cooperates, opening the charging port door. Tesla also interlocks the car so that you can't drive away while plugged in. I'm surprised that Tesla hasn't deployed those robot chargers yet.

[1] https://www.youtube.com/watch?v=3y_J7fg03fA [2] http://mashable.com/2014/01/29/robotic-gas-pump/#hDxoRZBZ5Zq... [3] https://www.youtube.com/watch?v=uMM0lRfX6YI

They did deploy one station, between SF and LA. They say it has barely got any traffic, so they aren't planning to build any more. Others have pointed out that building the station was a requirement to get some CA tax benefit.

There is no value in automated refueling - almost everywhere in the world drivers re-fuel their cars themselves, so there is no money to be saved by making this process automated. The only exception to this are some american states where you can't do it yourself by law.

Not really. There's not a lot of value in autonomous refueling when everything else about vehicle operation is manual.

By 2050 gas stations will disappear to be replaced by electric charging stations which will be automated.

I'd say rather than refueling, repairing. Right now there is a human in the truck who will do basic repairs, change a wheel, etc. If you have driverless trucks, they will need people available all over the country to service them on a short notice. Although admittedly these jobs could also be automated in the long term.

My thoughts:

Fleets of self-driving trucks will be on the roads worldwide by 2020.

Those same trucks will have a ‘delivery driver’ inside the cabin for at least another 10 years.

By 2030 all sales of new trucks will be self-driving From 2030 onwards a robot such as the latest generation Atlas will be in the cabin to handle deliveries.

By 2050 very few, if any, human couriers will be used, instead people will have new jobs coordinating the self-driving trucks, delivery robots and facilitation depots.

DHL recently hosted journalists, customers, and experts in the field of robotics at “Robotics Day” in their DHL Innovation Center in Troisdorf, Germany. The company says “Robots will be part of the future of logistics, and we’re excited to be on the ground floor of what that future cooperation will be like.” https://www.youtube.com/watch?v=YeL-YtaUkWc&feature=youtu.be

There was a story about a Tesla's radar being blocked by a moth, that smashed to its front. A fully autonomous vehicle would have had to stop and wait for someone to come by and clean the radar. Or there need to be lots of additional minor repair gadgets added to the car to clean sensors in case of dirt.

There are already lots of failure modes that require a tow truck to rescue a human-driven car. I bet the reduction in human-caused accidents (sleepy, drunk, distracted driver) will more than make up for weird new failure modes like sensor-splats.

>we'll see a rise in full-service highway gas stations to fuel driverless trucks

Not really - all AV trucks will be owned by consolidated corporations, think Uber for trucks. These corporations will have service centers in strategic location, AV trucks will come for refueling to these centers.

I am still yet to see an answer to the liability question on driverless trucks. Are the truck manufacturers going to take that insurance on? It cost Airbus billions of dollars when their software screwed up and they got massive amounts of government support to basically bail them out.

Presumable automated trucks will kill less people than human operated ones, so the insurance will be cheaper.

I am sure the companies who will operate those trucks and the manufacturers can figure a way to divide liability.

They seem to be doing just fine figuring out how to deal with the liability associated with the 3600 or so people who die in crashes involving big trucks every year.

Refueling of long haul trucks is already automated.

9% more unemployed is a recession. 38% more unemployed is a crisis/revolution (unless we play our cards right)

One for one replacement is a flaw, but also it doesn't take into account that efficiency often times leads to more work being done.

Like accountants. Excel and other software resolved a ton of their work. You didn't see a collapse in the number accountants. They just crunch more numbers, do more analysis.

uber made taxi's more efficient and I now ride in one 10X as often.

>We're seeing this in the more advanced law firms. What used to take a senior attorney, a few junior attorneys, a large number of paralegals, and a big clerical staff can now be done by one attorney, one paralegal, and a lot of software and databases. The OECD study claims that the risk to people with high levels of education is almost nil. Ask any newly graduated lawyer trying to get a job.

Software has barely put any lawyers out of a job. There have been staff cuts but that is more because babyboomer and older attorneys didn't know to word process. WordPerect (and now Word) are the reason legal secretaries are a dying profession. GenX and Millenial attorneys find it easier to it themselves.

Like accountants, a lot of this increased efficiency just leads to doing more work to make the end product better. You used to have research cases in books. This sounds inefficient, but it's actually not that bad. The book sellers summarize and catalog cases. You look up the issue and see what cases there are, read them and that is that.

Now, it's all done in digital databases. But it's not really faster. The digital databases still require a person to read the case, summarize, and catalog it. But it also allows you to read cases that weren't published. So it's now possible to find better cases, but you gotta spend the time reading it. So instead of spending an hour to get 10 cases sort on point, you have access to every case. But it takes well over an hour to get the perfect case. You get the better case, but you spent more time. Maybe you'd need a team of 50 lawers to find that case in the 1970s, but nobody would actually do that.

Ross--that startup claiming to be an AI lawyer--is spamming legal news with press releases vaguely claiming it can do all of that instantaneously. I sincerely doubt it. There is a reason they don't have a demo on their website.

The only software that I know that is doing what I'd consider legal work is the machine learning applied to document review--called predictive coding. But even that is only a response to the massive amount of information generated in digital offices. Workplaces produce many times more emails and documents than they did 40 years ago. E-discover exploded in the late 90's and early 2000's.

And even that work had already been outsourced to contractors who hired reviewers--often in India--to do the big projects. It's reduced some legal jobs, but only jobs that were created by technology. You can't apply the machine learning to smaller sets of documents since there isn't enough data to learn from.

The reason the legal market crashed in 2008 is because the M&A stopped and because those mortgage backed securities required a ton of legal work. And then the overall recession made companies sensitive to legal bills.

Newly graduated lawyers had a hard time finding a job even when the market for corporate lawyers was hot. Because tons of schools opened up law schools to cash in. They flooded the market with people who had no business being a lawyer. Even in 2005, the median law grad made less money than a programmer.

edit: wow I really ranted here. embarassing. I'm actually putting off doing document review because it's boring. I wish I had an AI that could do it for me.

I just realised that I plagiarised most of your points in another comment before reading yours.

The lawyer industry as it is, is I think a form of archaism. In most transactions there is zero justification to custom make a contract. Most sales of houses and businesses have exactly the same terms and the only reason you need lawyers to read and customize contracts is because other lawyers customized contracts previously and someone has to do the due dilligence of figuring out what they did. I am sure you could design a robust legal system with simple laws, where citizen do not need an expert to understand their rights and obligations, and where everyone would use standardised contracts for the quasi totality of transactions.

In fact it is a bit shocking that a common citizen cannot possibly understand the hundreds of thousands of pages of law and jurisprudence they he is required to comply with and needs to pay (dearly) an expert to do that for him.

What about neota logic ? they seem to build pretty capable software for legal.

Also, as for your claim that "works fills the available time", what about competition, new models of billing(fixed-fee), etc ?

I'm not sure what Neota Logic does. Looks like it's a best practices guide prepared by Litter Mendelson turned into software.

This could be a good example of advancements creating more work. My firm does a ton of employment law advising. My coworker in that department is always bitching about HR workers in her clients organization making dumb legal choices on their own. This software is designed for that sort of thing. The HR drones will instead use this software instead of their intuition. And I'm willing to bet this drives legal work when the HR drones run into issues that aren't clear (much of the law has no clear answer). Before the HR people wouldn't even know they had a legal issue to talk about.

Nobody was ever paying 500 dollars an hour to a firm to ask general compliance questions. In fact, compliance work is generally not considered legal work. They had an HR person making their choices before, not a lawyer.

Though I think my firm sometimes prepare best practice guides for clients. I'm not entirely sure if they are paying it, but I suspect they do. So maybe that software would reduce that sort of work.

>Also, as for your claim that "works fills the available time", what about competition, new models of billing(fixed-fee), etc ?

You'd think it would have a bigger impact, but we see this in many industries over a long time. Clothing got cheaper, and now we own a lot more clothes.

It's really shocking how much lawyers get away with charging based on how many unemployed and underemployed lawyers there are. I think it's partially because a lawyer is only worth something when they are experienced, and unemployed lawyers never get experience.

but also the value of legal work is really hard to objectively value. I can't even tell if the quality of work actually pays real world dividends.

>> I'm not sure what Neota Logic does. Looks like it's a best practices guide prepared by Litter Mendelson turned into software.

As i understand , it lets lawyers decode their logic/thinking into the software, so you get more than a best-practice guide(assuming best practice guides are like what we have in software) - because to build a full best practice guide that includes all the corner cases and the fine details, etc is just not practical(too much to read, you'll need an expert anyway).

But once you can gather information from the client(in a branching logic kind of way), you can create detailed advice that's very focused to the client's exact problem, and that usually offer much higher value - if you do it well.

>> (much of the law has no clear answer)

So many decisions in law are probability based ? the lawyer knows that doing X has 90% chance of being OK and 10% of being not OK (roughly) depending on corner cases or how laws will be interpreted in the future, etc - and builds a plan upon that, taking risks into account ?

Because maybe neota supports probabilistic logic, not sure. But for example i know some medical expert system software do us that.

>> Clothing got cheaper, and now we own a lot more clothes.

We probably consume more of most things, and yet employment in some sectors decreased.

>> I can't even tell if the quality of work actually pays real world dividends.

So selling is key. And probably risk reduction is key.Right ?

Computers have a pretty good track record doing risky stuff in a safer way and handling complexity, because humans make mistakes, and are limited with regards to complexity. It could be a big selling point once legal software matures.

I work in a bank and we regularly pay for legal opinions or advice on whether something we are thinking to do is legal or compliant with the laws and regulations. Are you referring to something else when you say compliance work?

It's ok Mr Rhino.

I come here primarily for the rants :-)

You learn interesting things and it sparks ideas.

Rant on Mr Rhino!

As someone constantly dealing with lawyers on a day-to-day basis, I cannot wait for the day that self-empowered "legal hackers" are able to compete against lawyers using tools such as Watson/ROSS.

Imagine being able to dictate grounds for relief, and have the machine draft the paperwork for you, ready to sign/witness and file.

Imagine not having to pay a lawyer $400/hour to answer what should be a simple question.

Imagine being able to have computer-based training about the acts/statutes you are currently referencing in your process.

I can't wait.

> one attorney, one paralegal, and a lot of software and databases

Now imagine what would happen if common law in the US was abanonded in favor of civil law, so that the data was a book of laws, rather than a history of cases...

>Ask any newly graduated lawyer trying to get a job.

Is this actually a global thing? I mean does this apply to smaller cities as well?

Sigh. It doesn't matter. People have this bizarre conception of the economy as an exact lock-fit between jobs and work that needs doing, and they have a hard time imagining changes to that fit, even though throughout human history there has been constant change.

In truth there is a nearly infinite amount of work that needs doing. We just get by with most of the work not being done. Just go back in time and think about all the work that wasn't being done before various jobs existed. There used to be a time before the video games industry existed, there was a time before the movie industry existed, there was a time before people could support themselves as authors or journalists or watch-makers.

Are we going to live in some sort of world where there's a pool of unused labor and people just sit around thinking "whelp, it's a shame there's literally nothing, no possible thing, for them to do, oh well"? Of course not, that's a fantasy that's based on a complete misperception of economics. No matter how much we automate there will always be work for people to do. The more important question is how equitably they'll be compensated for that work and whether or not our education system is adequate for the world as it exists.

That is all true. Still, try pointing that out to truck drivers (for example) when they start getting laid off because the company can't handle the automated competition. Economy-wide, there will always be jobs for people to do, it is the large number of individual workers who will be scared for their particular jobs.

This is true, and very important. Volatility in jobs can potentially put those at the bottom at an increased disadvantage, while consolidating the advantages of those who are already well off. Partly this is because people are "falling down" the experience ladder when starting a new career and their lack of economic buffer means they also can't negotiate decent wages. And too, new industries means that the hard won organizations (like labor unions) and established norms in pay and working conditions are no longer present, putting folks at another disadvantage.

Which has almost certainly been a factor in what's been happening to labor since the 1980s, who have seen their wages remain stagnant and job security fall through the floor (especially for those under 30) while the economy and productivity continues to grow and grow. This is, I would say, fairly obviously a bad thing, and requires a lot of rethink in order to makeup for the extreme power imbalances and iniquities at play, but very little of substance has been done about it so far.

The problem is not in the amount of job to be done. The problem is the cost of replacing a human being with a machine. Until the cost is higher than the worker's salary, that worker is fine. When it becomes lower, there is no job for him any more. So far we've always been in a situation where for any job that could be done by a machine, infinite more would only be doable by a real human. But we'll quickly reach the point where there will be NO jobs that can't be done by a machine for cheaper. On the other hand, if information technology keeps evolving much faster than robotics, the kast jobs to actually disappear will be the low qualification manual jobs. Note that truck driving is an intellectual job, not a manual one.

Sure, there will always be work for Bob to do. Clean the streets, fight crime, cure cancer, write bestselling novels. But there won't always be someone willing to pay Bob for that work, because there might be more effective options that don't involve hiring Bob. If you think Bob will always have a job due to some law of economics, I'd like to see that law spelled out. (Comparative advantage doesn't fit the bill, because it relies on the unrealistic assumption that capital is immobile.)

>No matter how much we automate there will always be work for people to do. The more important question is how equitably they'll be compensated for that work and whether or not our education system is adequate for the world as it exists.

And if it isn't, then those people who've been automated out of a job will find something to do, such as crime.

I think I finally get why libertarians have a boner, so to speak, for legalizing prostitution. That's the ultimate expression of adding elasticity to the labor market, and vastly more people could be participating in that market than do now. It makes such claims about there being infinite work to do ever so much more believable.

No, you don't "get" it. Libertarians just don't think it's the state's place to decide the conditions under which consenting adults have sex.

Forcing people to relate exclusively under free market rules is very much deciding the conditions under which consenting adults have sex.

The absence of government regulation is not the same thing as "forcing" anything.

Hm. The subtitle is "Destruction of large numbers of jobs unlikely, says new OECD Study", but the very first table gives a list of percent of jobs automatable by country. It ranges from 7-12%. That is a HELL of a lot of jobs. The argument is, "have no fear! it won't be 47% of jobs automated!" -- only 10%. 10% job loss is a labor crisis. Of course if you have all of those retrainable as other jobs (including automation supervisor) then maybe it won't be too bad, but that seems pretty unlikely and this article doesn't exactly quell concerns.

Something like 70% of Americans worked in agriculture in 1850, and like 5% did in 1950 (many of whom were probably migrant workers from Mexico).

We've gone through radical change before.

how it goes down is important though. if automation follows after labor begins to move to something else then that's one thing, but the reverse, automation putting people out of work, mean there might not be anything for them to realistically move to.

A booming business in automation and robot repair.

The whole point is that you need fewer humans in the loop. If every factory worker automated out of a job was hired back to work on the robots, we'd skip the automation part entirely.

Good luck learning to program or repair self-driving trucks when you've been a truck driver for your whole career.

Self-driving trucks have mechanical parts too. Even if you automate basic mechanical maintenance, operation and supervision will still require human intervention for a relatively long time.

One of the bigger issues now is you're competing with the entire world. When global shipping can get any product almost anywhere in a few days, economy of scale wins.

With oil becoming more expensive, for how long do you think global shipping will stay cheap? "Economy of scale" is many times an euphemism for "burn fossil fuels with subsidized costs".

We can go back to using sails. Cheaper but slower, so it'll work for a portion of shipping

Yes, and they where replaced with mainly useful jobs. But how many useful jobs can keep coming up with ? At what point does it just become sick extravagance of goods(materialism) and services we do not need?

Not as many jobs as were destroyed by the tractor. The tractor replaced over 70% of jobs, which was the proportion of farmers before the tractor.

"However, low qualified workers are likely to bear the brunt of the adjustment costs as the automatibility of their jobs is higher compared to highly qualified workers. Therefore, the likely challenge for the future lies in coping with rising inequality and ensuring sufficient (re-)training especially for low qualified workers."

Would seem that the 'low qualified' worker is the prevailing demographic. So that alone should make the risk, as an overall percentage, much higher than 10.

FWIW, my employer has been targeting all levels of workers for past several years. HR is gone, warehouse people down 65%, engineering down 40%, logistics down 30%, production down 35%. Mostly through process automation, some through robotics, and some through out-sourcing. And there has been a careful and selective 'firing' of customers. Both Net and Gross have significantly increased for three years. As for me, just sit here and write the code to do some of this. Am I evil?

Honestly, I read these studies and think nobody writing them has any idea of what these jobs actually entail. Driving a truck is not the be all and end all of being a truck driver. The driver covers a great many tasks, from admin and security, to safety inspections, to dealing with law enforcement and boarder guards. I see no talk of how the robot is to replace those functions.

I do see sweeping statements about rebuilding entire infrastructures in order to accommodate robots (ie doing away with roadside safety inspections or eliminating boarder crossings). Every day I see ships come into harbour. Creating an autopilot for a ship is simple compared to one for a trucks. Yet every day a speedboat heads out to deliver the harbour pilot, and tugboat captains stand guard as tankers approach the bridge. Automation has failed to remove any of those jobs. The engine room crews may be a little smaller than in days past but the bridge crews, the drivers, are still there. We wouldn't have it any other way.

And aircraft ... and trains ...

You could have truck drivers just at the start and end points of a route. Trucking companies can have representatives at the border, who then guide it through customs. No need to sit in the vehicle for hours.

Again, it isn't the physical act of driving that matters. It's that the truck driver is vetting the cargo. He is responsible for the contents, he is going to jail if there is something illegal in his truck. Someone jumping in the cab just shy of the boarder isn't going to be of much use to a boarder guard asking questions.

The actual study is a bit more detailed and it suggests some interesting things.

One is the evolution of work and jobs over time. As a pretty relevant example, the migration of shopping to "online" vs in store means two things; First you don't need the goods in a metro area, and second you need to get them to the customers. If you replace all current "shopping" traffic with "delivery" traffic it really cuts into the buyer's experience (waiting for delivery) but drone delivery allows the delivery component to scale. Of course drone delivery has to be reasonably local, but needn't be closer than a 10 miles or so, that means a "town" outside of the city that is well served by cargo container delivery can then provide the point where bulk delivery switches to individual delivery. That implies a economies with the movement of containers above and beyond the current system of trains and trucks. Either additional rail networks or lots more trucks. If those trucks are automatable, well that helps as well.

So if we imagine some "delivery only" roads where only robotic trucks are allowed, that lead to warehouses where end product dispersal is done, to smaller warehouses where local delivery can be queued/expedited. Walk that backwards to figure out the things you need in order to deploy that.

At which point it would be interesting to evaluate the energy efficiency of that system over the current one to make sure what you get back in efficiency by automation you don't spend on additional energy.

> we imagine some "delivery only" roads

This is possible but unlikely. There have been numerous plans in the past to build such large infrastructure. Like large pneumatic tubes to every house for deliveries. It usually turned out to be too expensive to build.

The trick though is to understand the economics of what you're trying to do before you implement it. Additional rail lines are quite expensive, one lane roads not so much.

Starting small, consider an autonomous truck that uses electricity like a city bus to move a container along a single lane protected from other traffic. You have one end in the port of Oakland (busy container terminal) and the other in a distribution warehouse outside Livermore (lots of open space, access to freeways and rail). Now one has to price out the cost savings of having containers appear automatically in a transhipping location without burning diesel or getting stuck in traffic.

You save a bit on the driver, you save more on the gas, and you take roads off the freeway which will help you to encourage local municipalities to give you easements for your autobot trucks to drive on.

It would be fun to sit down with a trans-shipper and find out their costs and whether or not you could save them time and effort.

Their conclusion doesn't follow from their numbers.

> Overall, we find that, on average across the 21 OECD countries, 9 % of jobs are automatable.

As in, right freaking now? Blimey, what about the next few decades, then? Not to mention that 9% is already a big number.

That being said, the more automation the better. We just need something better than punishing the jobless with poverty.

That is the concern. AI and robotics keep evolving, perhaps at an exponential rate, and therefore more and more jobs will be slated for automation.

The study should be thought of as valid only short-term.

OECD deemed wrong. Replaced by robots.

We have seen exactly the same fear with the appearance of machines and automation in the industry after the war. Look at historical news footage and people were concerned that this would lead to massive losses of jobs. And it did, the industry is a lot less labour intensive today, at least in the western world. But jobs were created elsewhere and the economy adapted.

Beside, with outsourcing to China, there aren't that many manufacturing jobs left in the US and Europe for robots to take anyway. I think where robots will have a huge potential is for domestic tasks: cleaning your home, taking parcels while you're away, preparing food, doing the laundry and ironing. A home robot that would do all that would have a huge market. But they are not going to replace an existing workforce, rather free up time for women (who still predominantly bear the burden of these tasks) and enable them to focus on their career.

I think there could be another wave of automation in the service industry, but it wouldn't be robots, just software. Today there are a huge number of manual tasks done in the service industry: processing invoices, preparing financial accounts, payslips, paying a lawyer to rewrite the same contract over and over, etc. Some of that can be automated by outsourcing it to a provider who has the means to automate it, but a huge fraction is just too specific, customized to a business, to justify paying for an IT team to build software for it. The way to automate it is for business people to build their own automation. In a way Microsoft Office has done partially that (an accountant with Excel has the productivity of 50 accountants from the 50s) but I think we can gain another order of magnitude of productivity by enable people to code. For that, basic coding skills should be widely deployed, in a very simple and highly productive language. And I am not convinced this language already exists.

Nitpick :

>OECD Working Papers should not be reported as representing the official views of the OECD or of its member countries.

And yet they always are.

Numbers wise, even if Frey and Osborne's supposedly apocalyptic numbers are correct that's 47% job destruction over 20+ years, or roughly 2.35% per annum. Since we already destroy between 10% and 15% [0] of jobs annually (depending on where you are and who's doing the measuring) this really doesn't represent that big a shift. Given some of this is likely due to automation anyway, the lower figures given in this paper would barely register.

[0] http://econweb.umd.edu/~haltiwan/c12451.pdf

A model of human work:

- The number of people occupied per industry is inversely proportional to the number of industries in existence.

- The barrier to entry is also inversely proportional to the number of industries in existence.

The process for every new industry generally goes something like this:

- Stage 1: mostly makers (100%)

- Stage 2: some makers (25%) + some operators of some tech that raises general efficiency (25%) (50% slowly moves to some new industry)

- Stage 3: few makers (5%) + some operators (10%) + some supervisors of some autonomous tech that maximizes efficiency (10%) (25% slowly moves to some new industry)

Was this process historically painless and ultra efficient? No! Did it do the job again and again with some acceptable level of efficiency? Yes!

The argument that this time is different this will stop working is getting stronger and stronger.

I personally think that is not the case. With some pain that require good policy, up and downs and so on this is the way we operate, we adapt.

Take prostitution (the oldest job in the world!), the next thing is robots yes, but also cam girls. Lower barrier to entry and maybe even bigger industry with lots of specialization.

And next thing for cam girls? Is of course virtual cam girls but also avatar builders/modelers/players even lower barrier to entry and potentially bigger industry and so on.

Let's see if a deep learning system outperforms OECD economists on this question.

Judging by the number of times I saw OECD economists getting any prediction right, I'd say a two headed coin will at a minimum even them.

Having looked at the assumptions used to get their results, I suspect dice might outperform the OECD on this one.

We should also be asking, who owns the robots? If companies like Foxconn go robo-conn and move to 70% automation [0], that's not only a million people being fired, it's a business owning an automated labor force that doesn't employ people, yet produces goods. If we can calculate how much labor these robots do compared to humans, one solutions would be for the government to tax the fuck out of robots to protect jobs and to compensate for rising social service costs (assuming the paranoia over job growth has merit). Of course, that's also what they could have done to counter the market dumping by overseas outsourcing, but they didn't give a fuck. Much automation has also happened in US factories a while ago with bottling and food and high-tech. To the government's credit though, we survived.


[0] http://www.computerworld.com/article/2941272/emerging-techno...

The most imminent threat to jobs that I can identify is drivers. Taxi's, trucks, delivery vehicles and buses.

We can see rapid improvement of self-driving technology. For sure, tech moves into mainstream more slowly than we generally imagine it will, but the sheer numbers of persons who could be displaced make even a relatively protracted implementation a big problem. Drive through Mountain View, Ca. on any given day and you'll see multiple Google autonomous vehicles. Uber, Apple, Tesla, GM, Volvo, .. most major auto companies, some minor companies and even non-auto companies are working on it with a high level of focus.

Two aspects of this movement may inhibit widespread deployment and slow the rate of robotic vehicles subsuming human jobs:

- The pace of legislation and the attending regulatory infrastructure. Governments are cooperating for the most part, likely attracted by the prospect of safer transportation in general. So I think this will not materially delay the roll out.

- The dilution of talent as these efforts compete for technologists may slow the progress of all of them, unless individual engineering leaders can attract and retain the top people.

The numbers look like this (2014), [1], [2], [3], [4]:

233,700 Taxi

665,000 Bus

1,797,700 Semi/Tractor-Trailer

1,330,000 Delivery

In 2014, that was roughly 4 million people who make a living driving. And the 2 most valuable companies in the world, Google and Apple, are working very hard to put these drivers out of work. What makes it particularly difficult is that these workers are typically uneducated and will have no place to go except minimum wage service jobs. Moreover, many former manufacturing workers took these driving jobs as a downgrade to income. And it looks like they have another target on their back.

Where do they go from here? Speaking for myself, I think the country has a moral obligation to make some accommodation and not just wave our hands a bit and say 'oh, they'll be fine'.

[1] http://www.bls.gov/ooh/transportation-and-material-moving/ta...

[2] http://www.bls.gov/ooh/transportation-and-material-moving/bu...

[3] http://www.bls.gov/ooh/transportation-and-material-moving/he...

[4] http://www.bls.gov/ooh/transportation-and-material-moving/de...

Moving people and things is way bigger than just vehicles on public roads, too. There are giant earth-movers in pit mines, metro trains, and other fairly car-like examples. There are also less obvious examples like the increasing automation of Amazon warehouses: where once pickers moved amongst the shelves in a warehouse, now pickers are stationary and robots move entire banks of shelves around, entirely automated, and the same amount of picking and packing is done with fewer pickers.

These are all facets of the same problem, and it's everywhere.

Find it entertaining that a GLM is the best method to detect whether a job or job class is at risk.

GLM (global free market) having not read the article ... but the risk of job loss to automation is even more dire in those GLM labor countries. We have already seen jobs migrate from US -> China -> Vietnam -> Sudan. And each step the level of automation increases. Everything that happens in Japan will happen to the rest of the world at some point. Japan is the Canary.

The world has already passed peak job. Our task as programmer has always been to automate everything, including ourselves.

Someone will soon build a totally modular, reconfigurable intelligent factory. An autofac, power and raw materials go in and anything and everything comes out the other side. For large classes of goods, this is an engineering problem, not a science problem.

Nobody has commented on that Russia only has 2% of jobs that they see are easily automated. That feels utterly ridiculous

OECD Study recreating easy says Neural Network

I'm just going to have to build better robots then, aren't I?

I'd say it's inevitable. The actual important question is when.

5 years?

So it will happen around AI beating humans at go?

(Prediction^HopeItWontHappenToMe) * Scottyfactor = Real Timeframe

Which happened earlier this year.

9% is a lot. Globalization is the biggest problem though. Also, I think, is the realization that consumption is destroying the planet. This leads to higher consumption taxes (thank god) and more bartering via things like craigslist.

This piece is utter trash founded on baseless assumptions.

>Arntz, et al. argue that the estimated share of “jobs at risk” must not be equated with actual or expected employment losses from technological advances for three reasons.

So in other words, the title is false. Actual employment loss is what is at issue here.

>The utilisation of new technologies is a slow process, due to economic, legal and societal hurdles, so that technological substitution often does not take place as expected.

The first ridiculous assertion. Jobs wont be destroyed because it will take a long time to destroy them!

>Even if new technologies are introduced, workers can adjust to changing technological endowments by switching tasks, thus preventing technological unemployment.

Asinine assertion #2. People will just get new jobs when robots take their jobs! (of course what new jobs will be available is left unsaid).

>Technological change also generates additional jobs through demand for new technologies and through higher competitiveness.

Asinine assertion #3, virtually the same as #2. What new jobs will be created by the automated vehicles that put tens of millions of drivers out of work?

This article reads very much like the tripe offered by those who continue to argue that NAFTA and related "free trade deals" are actually good for workers.

It's a tedious and unproductive article.

There's real, interesting stuff to be said about automation and substitution effects - it is not to be found here. Yet another piece parroting the claim that displaced workers find new jobs adds nothing to the conversation. Look up any of the Tech Review debates, or if you're truly interested check out the pieces the OECD incorrectly claims to debunk.

'Asinine' is exactly the word, although I might choose "assuming the consequent" if I'm allowed to use a phrase. When the entire debate is over whether this time is different, whether new jobs will actually become available, it's impressively dull-witted to dismiss the argument by assuming the desired result.

I don't think the assertions are that ridiculous. Look at all the automation that we've already developed. "In 1870, almost 50 percent of the US population was employed in agriculture.[16] As of 2008, less than 2 percent of the population is directly employed in agriculture."

Does anyone today feel as though we've lost half of our jobs to mechanized farming? Not really, it allowed other industries like entertainment to flourish. Now that less of us have to do X, we can do more Y. I think it's absolutely a valid point to say "job loss will be slow enough that people will be able to adapt."

I honestly don't get the "robots will take our jobs" thing. It's sort of like "with more technology people will work less" - which has thus far proven to not be the case. People will always find more work to do, more ways to convince other people to spend money (and since the sales of X are now supporting less people, more money is freed up for Y).

The number of "jobs" available to a given population isn't really a function of the level of automation. It's a function of who is in control of capital/wealth and how that wealth flows between entities. Thanks to the automation we've already developed, we can already technically afford to provide everyone (in the US) with a basic income for the "job" of simply existing. Or we could place a high(er) tax on top earners and use that to double the minimum wage and split those jobs in half. Boom, look at all those new jobs created simply by a policy change. We just have collectively not chosen to do those things. Jobs are a red herring. We don't "need jobs" that are being destroyed by automation. The automation is adding value, not destroying it. We just have a setup where a small number of people are capturing most of that value. What we need are socioeconomic policies that even out that distribution.

>Does anyone today feel as though we've lost half of our jobs to mechanized farming, those jobs were "destroyed?" Not really, it allowed other industries like entertainment to flourish. Now that less of us have to do X, we can do more Y. I think it's absolutely a valid point to say "job loss will be slow enough that people will be able to adapt."

Someone else phrased it best, I read it from an HN user but they may have been using someone else's argument. Paraphrasing here:

"Machines made physical labor easier, requiring fewer people to do physical labor. Those people moved to doing mental labor. Now machines are replacing mental tasks too. Where can people go when fewer of them are needed for mental tasks?"

The last remaining type of labor would be creative labor. Writing, music, performance. Things machines either still struggle at or won't be able to perform until we have humanoids that move fluidly. Even then, humans seem to prefer humans performing "artsy" things - based on criticisms I've seen of robots who create art (mostly music and drawings).

Art also, notoriously, doesn't pay very well for all but a small fraction of artists. What happens when this market is over-saturated because it is the only job left for large parts of the population?

The underlying point is that not everyone needs a job. "Everyone has a job" is not an important fundamental requirement of functioning society. Everyone needs food, water, and shelter. In the early days, everyone needed a job because that's what it took to provide those basic needs. As we develop automation that allows those things to be provided to more people with less human labor, we shouldn't say "oh shit, we're running out of jobs, and people need jobs!" We should say, "how can we run a society where less human labor is necessary?"

What happens when this market is over-saturated because it is the only job left for large parts of the population?

I don't know, what happens? We've reached the endgame, we've fully automated all of the work required to keep people's needs taken care of, but we have some sort of problem because not everyone has a "job"? It's like some kind of dark comedy..."let's break all these robots that are growing food and building houses for us, we need jobs god dammit!" We just need to get away from the idea that "having a job" is a necessary goal in and of itself.

>The underlying point is that not everyone needs a job. "Everyone has a job" is not an important fundamental requirement of functioning society.

The discussion was about whether or not jobs have been lost. Not whether or not jobs will be or are necessary. For the time being, jobs are both being lost and necessary. At least for most people and people who don't enjoy being homeless.

In 1870, nearly half of the workforce worked in agriculture. Almost all of those "jobs" were "lost" over the next century. Did that have a meaningful correlation with unemployment rate?

There is not some fixed amount of "work that needs to be done" that in turn creates a fixed (and steadily diminishing) pool of jobs to draw from. The amount of jobs available is a purely socioeconomic function of who is willing to pay whom for what.

Take Facebook. Do the jobs there carve out some section of the total amount of work that needs to be done today? Not really, a guy made a thing, and convinced people to give him an arbitrary amount of money, and decided to give an arbitrary number of people arbitrary amounts of that money to help him out. A certain number of jobs were created because of the way people decided to distribute their money. Not because they took some of the limited "available job" slots.

> Asinine assertion #2. People will just get new jobs when robots take their jobs!

There's something to be said about this.

If there is unmet demand for something those people can do [1], and automation makes them more productive, and does not completely replace them; then reducing the price of those people work (by increased productivity) will make more jobs appear.

Note there are 3 "if"s there. I doubt many workers will see those three met for them.

[1] That means, people want more of it. Economists normally take this fact from granted, but it is not.

The hard question is not whether certain jobs will get automated out of existence--they will.

The hard question is why that automation will not result in the creation of new industries, which themselves will create new jobs. This has happened throughout human history. To say that won't happen anymore is a high bar.

That's why the pace of change does matter. If jobs are destroyed much faster than they are created, then there might be social unrest in the lag.

There's only so many times you can retrain a human. Plus, the "re-trainability" rate decreases dramatically with age.

In a decade or two there will be a heck of a lot of middle-age people hopelessly outpaced by the rate of change. 10 years after that, even young people won't be able to keep up.

While there are age-related declines in certain cognitive abilities, in the absence of conditions like neuro-degenerative disease, it doesn't mean that older individuals are incapable of learning new tasks and skills.

Differences among age groups are greatest comparing people in their 20's vs. 60's (or beyond). Even then, differences are modest.

See this article for more explanation: Clark R, Freedberg M, Hazeltine E, Voss MW. "Are There Age-Related Differences in the Ability to Learn Configural Responses?" http://www.ncbi.nlm.nih.gov/pubmed/26317773

There's also evidence of alterations in "learning style" with age but that's yet another consideration.

Why should this new industry's be save from automation? Yes, there is a tendency to create more "unique" handcrafted items - aka AI cant do unique Art like humans want- but the moment the demand for one piece surges, robotics will be there to take over replication.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact